Forecasting Spatially Dependent Origin and Destination Commodity Flows

Abstract

We explore origin-destination forecasting of commodity flows between 15 Spanish regions, using data covering the period from 1995 to 2004. The one-year-ahead forecasts are based on a recently introduced spatial autoregressive variant of the traditional gravity model. Gravity (or spatial interaction models) attempt to explain variation in N=n2 flows between n origin and destination regions that reflect a vector arising from an n by n flow matrix. The spatial autoregressive variant of the gravity model used here takes into account spatial dependence between flows from regions neighboring both the origin and destinations during estimation and forecasting. One-year-ahead forecast accuracy of non-spatial and spatial models are compared.

Keywords: gravity models, Bayesian spatial autoregressive regression model, spatial connectivity of origin-destination flows

JEL Classification: C11,C23,C31